Re: [bcause] PFCP not equal to FMC, Why not choose PFCP

["Wadhwa, Sanjay (Nokia - US/Mountain View)" <[e-mail]>]

I am amused to see the relentless effort here. First, based on Wikipedia PFCP would not work, then number of pages of the spec made it prohibitive, now it is "PFCP is defined by 3GPP, so it is not suitable".
But then you also suggest you could start with CUSP day one, throw it day two and move to PFCP. Why should we not start with the right interface day one, especially if it can provide a single solution irrespective of access.

BTW, PFCP includes both the interface to exchange state between CP and UP and the interface to pass in-band control e.g. PPPoE/DHCP signaling between UP and CP. You don't need VxLAN shown in your picture.
The third interface is completely orthogonal and nothing to do with dynamic state between CP and UP. It is a local management interface where CP provide single point for management and control for a set of UPs. You can think of it as a way to provide CLI on the CP for entire CUPS system. It could be NETCONF or gNMI or your favorite one, as long as you have a well-defined data model.

You talk about additional functions in BNG. With PFCP, you don't need to throw away the underlying protocol if you want to integrate another function in the BNG. PFCP (via PDRs) are flexible in specifying forwarding between different access and network facing encapsulations i.e. [access <-> network] as needed by different functions e.g. for fixed-access this may be [L2 <-> IP], for fixed-wireless [GTP-u <-> IP], for hybrid-access both [L2, GTP-u <-> IP], for (T)WAG [L2oGRE <-> IP or GTP-u]. Hope you get the idea. You also have the flexibility on the transport encapsulations in the access for fixed-network. Could be VLANs, MPLS-PW, VxLAN, L2oGRE, pick your favorite. No change to the underlying protocol.
But wait...what a tragedy that PFCP is defined by 3GPP.

-Sanjay

From: bcause <[e-mail]> On Behalf Of Songjun (Network)
Sent: Friday, March 29, 2019 6:24 AM
To: [e-mail]
Subject: [bcause] PFCP not equal to FMC, Why not choose PFCP



1.       PFCP is not important for FMC

[cid:image001.png@01D4E5F9.B419E3C0]

Mobile core is keeping evolution. PFCP must follow on it.



The CUPS interfaces are different between mobile core and BNG.

[cid:image002.png@01D4E5F9.B419E3C0]

vBNG-UP and vEPC-UP will run in VM or Container independently. We can do FMC now basing on large-scale deployed protocols (PFCP and S-CUSP), and keep smooth evolution following the development of mobile network and fixed network.

[cid:image003.png@01D4E5F9.B419E3C0]

We can guarantee there is PFCP in 6G and 7G. The latest information from 3GPP, they are discussing to change the mechanism between UPF and SMF.

Using PFCP is not equal to implement FMC.



2.       Why not choose PFCP
Why 3GPP not reuse Openflow?

?  PFCP use the openflow mechanism, openflow was a mature protocol then.

?  3GPP has its own requirements, using UDP for high throughput.

?  Defining mobile standard in ONF makes no sense.
Why we did not choose PFCP?

?  In 2016, we studied Openflow and PFCP, and discussed with PFCP editor (Peter from Huawei).

?  We found that nearly all requirements of BNG were needed to extended in current protocol. That seems to rewrite a new protocol.

?  Reusing Openflow and PFCP, the CUSP will be heavily affected by ONF and 3GPP. Discussing BNG in ONF and 3GPP makes no sense.

?  We decided to define a new protocol with good points from Openflow and PFCP in IETF for BNG.

?  Mobile core can not use PFCP for interworking with different vendors now in technical reason.

?  When talking about using PFCP, let me recall the story of T-MPLS(MPLS-TP).  IETF asked ITU-T to stop using MPLS or standardize T-MPLS in IETF.  ITU-T lost the control of T-MPLS, and the standard process was delayed.


3.       DBNG story in BBF

?  I attended last two BBF meetings.  I found that they want to put everything into MS-BNG, as the following picture. According to Nokia's IETF Draft, EPC and 5GC will also be input into it.

?  If just put all function together without any changes of outside interfaces and protocols, it is an implementation issues, not a standard issues. We can do it now because we have all related products.

?  If needing converging functions and changing protocols, someone said that the work could be finished in 5 month.  I do not believe that it can be finished based on running code instead of paperwork.

?  There are about 20 people attend ATA meeting, with more than 1/3 from Huawei and ZTE.


[cid:image004.png@01D4E5F9.B419E3C0]
BTW, this picture was rejected in BBF. It is not easy to draw a clear picture including all functions, even to me.


4.       Suggest IETF  start CUPS protocols for BNG

?  CU separation is an internal mechanism, unaware to outside, without adding new functions and changing current protocols. TR-384 shows how to separate the functions into CP and UP, it is enough to start the protocols design, basing on running code and scale deployment.

?  Most BNG in current and in near future, will be used for fixed network users. Requirements on fixed BNG is clear and urgent. How to do FMC is still in discussion, with multiple ways to support it.

?  S-CUSP's running code is 1/10 of PFCP, because of the different service requirement. CUSP for BNG is a very lightweight protocol, because fixed user does not migrate, always on line, very simple billing. BNG-UP can still forwarding, without BNG-CP. Mobile core reverses. Of cause, we can absorb the good points of Openflow and PFCP.

?  IETF is the right place to define a protocol for BNG, 3GPP is the right place to extend PFCP.

?  If using PFCP and including unclear FMC functions, we will mesh 3 SDOs together, many device functions(BNG,AC,TWAF,HAG,EPC,5GC) together. When can we meet the real IP network requirements?





5.       How to deal with FWA requirements, such as HAG?

?  HAG can be deployed independently or with BNG. CUPS HAG can be added in WT378, with its owner CUSP..

?  Venders can provide independent HAG or integrated HAG with BNG according Carrier's requirements.

?  Optimization on integrated HAG is a kind of vendor's implementation, not standard issues.

?  CUSP for HAG can be defined according to HAG requirements, with smooth independent evolution.


Simple and Flexible are the keys for success of IP network, when you think about Ethernet/IP against ATM.